ZoneMinder Forums

ok then i think i am getting somewhere on this i can have all 16 of the composite cams working lovely
load is about 2 (with a 16 way remote view of zm4ms on site)

as soon as i add 1 ip cam (axis 206) load goes up a little and another 5% processor used all is fine if i add another i get huge jumps in load

could this be a network card issue??

Hi James,

You aren't by any chance throttling the fps on the network cams at the zm end are you? Instead of in the cgi string passed to the cam?

well i was using mjpeg streams from the cams limited to 2 fps and unlimited at the zm end.

Whilst clutching for straws i have changed all the paths to jpg and set the limit on the zm end (which i assume is correct for jpeg i understand why the limit needs to be at the cam for mjpeg)

following on on my theory this is network related i am using the built in network card, could this be my issue if i put a decent pci network card in would this help?

this definetly something to do with the netwrok cams
so i suppose its either
a the network card in the mobo
b the fact thats its decoding the jpgs before analysis is just to much for it
c something else network related

Network cameras are definately more CPU intensive than regular ones, the process of decoding jpeg images really is a hog and can end up being what is using the majority of the CPU.

Phil

I should add that 1.22.0 will have some experimental jpeg optimisations that might speed things up but to what degree I don't know.

Phil

thing is id understand that if i was at 0% idle and each zmc/zmc for that monitor taking all the cpu but they take between 3 and 6%. and i end up with a 25-35% idle on the processor. But a load of 22

Im assuming these built in mobo network cards make the processor do all the work.

I also have zm4ms on this site but zms is taking 0.4% cpu thats all.
But obviously networlk bandwidth is high and im assuming this is th eproblem but i dont know.

16 composite cams load 1.2-1.4
5 network cams load 23

I cant see thats right

James

ooohhh experimental optimisations
mmx?
Cant wait (well i normally cant why should this be any different lol)

Your load does seem high but don't underestimate the proportion of time jpeg handling does take! This is one of the reasons I am so keen to push raw images to your app when it's on the same network segment.

The optimisations aren't mmx I'm afriad (that's much too platform specific to put in a general release) but more to do with caching quantisation tables and such like. Don't get too excited! There are also many changes to the zone handling, for instance they can be any shape now and not just quadrilateral, and I've optimised the motion detection a little bit around that as well.

Phil

i was a bit scared of this raw format but i think you may be right, but zms takes less than 2% of the proc streaming 36 images so where does the load show up. I would haved thought that zms would be taking tonnes of proc time if its that heavy?

I've never really been able to get 'top' to add up, though it is useful for relative ordering. I suspect perhaps we need some more sophisticated profiling and performance tools to really figure this out.

Phil

do you think i may be close with the network card or am not even in the right forest

No idea mate, I'm afraid but it sounds possible. Is it 100Mbs or 1Gbps?

Phil

100Meg but i cant seem to find anyway of checking the status of the network card ie its throughput etc. I assume its like a disk to the kernel and will wait for stuff. I was wondering if i could tell it to priortise ceratain traffic. Assuming it is the netwirk card. Im ganno do some tests with another machine and a 3com pci card to see if im close. Its all part of this huge learning curve, im hoping in about 10 years it will be a little less steep lol